Edit Article

Years ago, I was building MEMS systems where sensors and integrated circuits had to coexist on one chip.

The hardest part was "teaching" the chip through low-level firmware. Over I²C, we streamed sensor data, watched how the device actually behaved, and wrote calibration parameters back into it, loop after loop.

But the real work was never just the firmware on the chip.

It was the automated validation environment we had to build and keep running. Test benches operated 24/7 for months, combining multiplexers, programmable power supplies, precision Druck pressure controllers and pumps, environmental chambers, refrigerators, ovens, temperature controllers, data acquisition systems, oscilloscopes, and whatever other measurement instruments we needed. That entire stack existed to characterise one tiny piece of silicon across a huge matrix of operating conditions.

Once we had collected enough data, the real engineering started.
We analysed enormous datasets using engineering equations, regression, statistical methods, repeated calibration runs, and countless iterations to extract the right parameters. Reaching production-ready calibration took months — sometimes years.

Today, AI and modern statistical computing can evaluate millions or even billions of possibilities in a fraction of that time. Regression still uncovers relationships, Monte Carlo still explores uncertainty through simulation, and Bayesian inference still refines probabilities as new evidence arrives.

The engineering principle, however, is unchanged.

The most effective production systems don't let AI handle every execution. They use AI to discover the optimal solution, validate it, and then bake it into deterministic processes executed by rules engines, parsers, and specialised subsystems.

AI accelerates discovery.
Deterministic systems ensure consistency.
Engineering discipline remains the foundation.
Technology has changed. The engineering mindset hasn't.